AI Guidelines Draft Content Map¶
This page tracks migration of GDD Group AI Use Guidelines, version 1.0, August 2026. The source is a working document, so inclusion here does not mean the policy has been approved.
| Source content | Handbook destination | Treatment |
|---|---|---|
| Purpose and living-document statement | AI overview | Migrated as principles and review expectations |
| No identifiable or sensitive data | AI landing page; Research data | Migrated as a core GDD requirement |
| Tool access to directories | AI landing page; Research data | Migrated as permission minimisation |
| Identifier, quasi-identifier, and metadata examples | Research data | Migrated and organised by category |
| Minimal and dummy datasets | Research data | Migrated as an operational workflow |
| Aggregated datasets | Research data | Migrated with publication-readiness, small-cell, and complementary-suppression checks |
| Human-written first draft | AI landing page; Scientific writing | Migrated as a GDD policy choice |
| AI-assisted coding and tests | Coding & analysis | Migrated with accountability and validation controls |
| Authorship and attribution | Disclosure; Scientific writing | Migrated |
| Transparency in methods | Coding & analysis; Disclosure | Migrated |
| Trainees and students | AI landing page; Coding & analysis | Migrated; link to supervision guidance later |
| Funder and journal compliance | Disclosure | Migrated; claims about named funders need checking |
| Field and clinical AI tools | AI landing page | Explicitly placed outside the guidance’s authorisation |
| Administrative uses | Not yet migrated | Needs a confidentiality and meeting-recording rule |
Decisions still needed¶
- Name the policy owner and escalation route; replace references to individual people with durable roles, while listing current contacts separately.
- Decide whether the prohibition covers all AI systems or specifically externally hosted generative AI; define any institutionally controlled exception explicitly.
- Confirm whether the draft small-cell threshold of [5] should become the approved minimum or vary by study and context.
- Decide whether prompts and outputs must be retained for substantive analytical use and, if so, where.
- Resolve the unfinished proposal about marking AI-authored commit messages; a repository-wide rule is not yet included.
- Define what counts as a “first draft” for outlines, translations, repeated boilerplate, and collaborative rewriting.
- Add a rule for confidential meeting notes, recordings, email, translation, and other administrative use.
- Verify current UNIGE, journal, and named-funder requirements before citing them.
- Decide approval authority, effective date, review interval, and how changes will be communicated.